Spurious 'max ice volume' for South hemisphere causes output conversion error in ice_diagnostics on gbox128

[phil-blain]

I got another output conversion error, this time in CICE. In contrast to CICE-Consortium/Icepack#378, I'm not sure what's the best way to fix this one. It happens for these tests in the base_suite:

• *_*_restart_gbox128_4x2_boxnodyn_debug_short
• *_*_smoke_gbox128_4x4_boxrestore_debug_short

Here is the backtrace from GDB (same for both):

(gdb) bt
#0  0x0000152d8b3e637f in raise () from /lib64/libc.so.6
#1  0x0000152d8b3d0e7e in abort () from /lib64/libc.so.6
#2  0x0000000001bf05d0 in for.issue_diagnostic ()
#3  0x0000000001bf61db in for.signal_handler ()
#4  <signal handler called>
#5  0x0000152d8b3e637f in raise () from /lib64/libc.so.6
#6  0x0000152d8b3d0db5 in abort () from /lib64/libc.so.6
#7  0x0000000001bf05d0 in for.issue_diagnostic ()
#8  0x0000000001bef838 in for.io_return ()
#9  0x0000000001c475bf in for_write_seq_fmt_xmit ()
#10 0x0000000000624259 in ice_diagnostics::runtime_diags (dt=3600) at /.../cice3/cicecore/cicedynB/analysis/ice_diagnostics.F90:943
#11 0x000000000041bf0d in cice_runmod::ice_step () at /.../cice3/cicecore/drivers/standalone/cice/CICE_RunMod.F90:378
#12 0x0000000000419d11 in cice_runmod::cice_run () at /.../cice3/cicecore/drivers/standalone/cice/CICE_RunMod.F90:83
#13 0x000000000040de40 in icemodel () at /.../cice3/cicecore/drivers/standalone/cice/CICE.F90:49
(gdb) f 10
#10 0x0000000000624259 in ice_diagnostics::runtime_diags (dt=3600) at /.../cice3/cicecore/cicedynB/analysis/ice_diagnostics.F90:943
943             write(nu_diag,900) 'max ice volume     (m) = ',hmaxn,  hmaxs
(gdb) p hmaxn
$1 = 1.9962409428228371
(gdb) p hmaxs 
$2 = -1.7976931348623157e+308

hmaxs comes from this global max earlier in runtime_diags:

CICE/cicecore/cicedynB/analysis/ice_diagnostics.F90

Lines 372 to 374 in 2b85126

	! maximum ice volume (= mean thickness including open water)
	hmaxn = global_maxval(vice, distrb_info, lmask_n)
	hmaxs = global_maxval(vice, distrb_info, lmask_s)

So I'm not sure how we get this spurious -1.7976931348623157e+308 for the South hemisphere... maybe lmask_s is not correct for box128 ? because vice itself seems to be correctly zero-initialized on the whole grid:

CICE/cicecore/cicedynB/general/ice_init.F90

Lines 2352 to 2359 in 2b85126

	!$OMP PARALLEL DO PRIVATE(iblk,it,i,j)
	do iblk = 1, nblocks
	do j = 1, ny_block
	do i = 1, nx_block
	aice(i,j,iblk) = c0
	vice(i,j,iblk) = c0
	vsno(i,j,iblk) = c0

I did not have time to dig further...

[eclare108213]

I've seen these formatting issues but not worried about them since they weren't causing the code to crash. The box tests don't produce data for both hemispheres, in general, and so the standard diagnostics might not be defined. There should be a way to initialize things so that the output can be formatted, even if it's not meaningful, or limit the output to only the region being computed.

[eclare108213]

This issue was addressed in apcraig#57